Mixtral 8x22B needs ~99.9 GB VRAM. NVIDIA H20 96GB has 96.0 GB. With Q4_K_M quantization, expect ~63 tok/s.
Operating mode
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
3.9 GB over capacity — needs offload or smaller quantization
Fit status
Runs with offload (needs ~3.4 GB host RAM)
Decode
62.5 tok/s
TTFT
3098 ms
Safe context
4K
Memory
99.9 GB / 96.0 GB
This setup is broadly balanced for this model.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | B | Runs with offload (needs ~1.9 GB host RAM) | 64.3 tok/s | 1642 ms | 4K |
| Coding | B | Runs with offload (needs ~3.4 GB host RAM) | 62.5 tok/s | 3098 ms | 4K |
| Agentic Coding | B | Runs with offload (needs ~6.1 GB host RAM) | 59.1 tok/s | 4763 ms | 4K |
| Reasoning | B | Runs with offload (needs ~3.4 GB host RAM) | 62.5 tok/s | 3661 ms | 4K |
| RAG | B | Runs with offload (needs ~6.1 GB host RAM) | 59.1 tok/s | 5953 ms | 4K |
Inference speed
Estimated decode speed (tokens/sec) for Mixtral 8x22B at Q4_K_M across popular GPUs and Apple Silicon, including multi-GPU rigs, using the fastest local runtime per device. Fastest is Mac Studio M3 Ultra 256GB at ~14 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.
| GPU / Mac | Memory | Quant | Speed (tok/s) | Fits? |
|---|---|---|---|---|
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 13.5 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 9.2 | Heavy offload |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 8.7 | Heavy offload |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 6.8 | Heavy offload |
2× RX 7900 XTX 24GB | 48 GB | Q4_K_M | 5.0 | Too big |
 RTX 5090 32GB | 32 GB | Q4_K_M | 3.9 | Too big |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 3.7 | Too big |
| 2× RTX 4090 24GB | 48 GB | Q4_K_M | 3.4 | Too big |
| 2× RTX 3090 24GB | 48 GB | Q4_K_M | 2.9 | Too big |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 2.6 | Too big |
| 4× RTX 3060 12GB | 48 GB | Q4_K_M | 2.6 | Too big |
| RTX 4090 24GB | 24 GB | Q4_K_M | 2.5 | Too big |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 2.5 | Too big |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 2.4 | Too big |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 2.3 | Too big |
| RTX 3090 24GB | 24 GB | Q4_K_M | 2.1 | Too big |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 2.0 | Too big |
| RTX 4070 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.0 | Too big |
Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.
How Mixtral 8x22B (141B params) fits at each quantization level on NVIDIA H20 96GB (96.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 55.0 GB | Low | B61 |
Q3_K_SBest for your GPU | 3 | 69.1 GB | Low | B61 |
NVFP4 | 4 | 79.0 GB | Medium | F0 |
Q4_K_M | 4 | 86.0 GB | Medium | F0 |
Q5_K_M | 5 | 101.5 GB | High | F0 |
Q6_K | 6 | 115.6 GB | High | F0 |
Q8_0 | 8 | 150.9 GB | Very High | F0 |
F16 | 16 | 289.0 GB | Maximum | F0 |
Copy-paste commands to run Mixtral 8x22B on your machine.
Run
ollama run mixtral:8x22bOpciones de mejora
NVIDIA H200 141GBOpción económicaSube la velocidad estimada de decodificación alrededor de un 56%.
Añade margen de memoria para más contexto y para que el modelo envejezca mejor.
~$30,000 MSRP
NVIDIA H200 PCIe 141GBMejor relación calidad-precioSube la velocidad estimada de decodificación alrededor de un 56%.
Añade margen de memoria para más contexto y para que el modelo envejezca mejor.
~$30,000 MSRP
NVIDIA B200 180GBMejora NVIDIASube la velocidad estimada de decodificación alrededor de un 160%.
Añade margen de memoria para más contexto y para que el modelo envejezca mejor.
~$30,000 MSRP
Yes, NVIDIA H20 96GB can run Mixtral 8x22B with a B grade (Runs with offload (needs ~3.4 GB host RAM)). Expected decode speed: 62.5 tok/s.
Mixtral 8x22B (141B parameters) requires approximately 99.9 GB of memory with Q4_K_M quantization.
The recommended quantization for Mixtral 8x22B is Q4_K_M, which balances quality and memory efficiency.
On NVIDIA H20 96GB, Mixtral 8x22B achieves approximately 62.5 tokens per second decode speed with a time-to-first-token of 3098ms using Q4_K_M quantization.
For coding workloads, Mixtral 8x22B on NVIDIA H20 96GB receives a B grade with 62.5 tok/s and 4K context.
On NVIDIA H20 96GB, Mixtral 8x22B can safely use up to 4K tokens of context. The model's official context limit is 66K, but available memory constrains the safe maximum.
Buy headroom, not only minimum fit. A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/mixtral-8x22b-on-h20-96gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: